Tag: machine-learning
All the articles with the tag "machine-learning".
The Mirror Problem
Published: at 03:44 PMA model trained on its own output loses range generation after generation, the tails of its distribution slowly disappearing. I hit this kind of loop in my own systems, and it changed how I think about the difference between a model learning and a mirror reflecting.
The Forgiveness That Dies First
Published: at 01:42 PMmuP makes learning-rate transfer work by keeping the set of good rates wide. I measured what low precision does to that forgiveness, and the surprising part: the transfer survives. The forgiveness is what dies first.
Making AI Agents Think Before They Retrieve
Published: at 09:00 PMEvery memory-augmented agent runs the same pipeline: query, retrieve, answer. Nobody asks whether retrieval should happen at all. Here's the missing cognitive control plane that turns memory access from a reflex into a decision, with controller architectures, a formal definition, benchmarks, and working code.